Application of machine learning algorithms to forecast prices of Sardinella brasiliensis commercialized in a South American supply center

Abstract

Abstract Small pelagic fishes develop important role in human nutrition especially in emergent countries which are considered an affordable source of protein ensuring food security, and with its fishery being source of income for several populations around the world. Despite fish nutritional composition present several benefits for human health, prices are pointed as the main factor to choose seafood as components of diet, highlighting the relevance of the economic analysis of these items once disturbances in its prices might alter the feeding patterns of populations worldwide. This study aimed to analyze the Brazilian Sardine (Sardinella brasiliensis) prices dynamics in one of the main markets of northeastern Brazil, evaluate possible reasons for its peaks and use machine learning techniques to forecast its future prices. The dataset used was obtained in the Pernambuco Supply and Logistics Center (PSLC) website, which contains a historical series of sardine’s prices from 2013 to 2022. The dataset was divided in train and test sections, the train section modelled using the Fbprophet library and a long-short term memory neural network in order forecast the future prices, then the test dataset was used to evaluate the predictions based in the root mean square error, mean absolute error and mean absolute percentage error metrics. Both algorithms reached low error metrics in its forecasts, however LSTM predictions were significantly better presenting lower error metrics than Fbprophet, showing their usability in the economic context of marine sciences opening the door to further studies of the dynamics of food prices around the world.